Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

LEAVES: Learning Views for Time-Series Biobehavioral Data in Contrastive Learning

Created by
  • Haebom

Author

Han Yu, Huiyuan Yang, Akane Sano

Outline

This paper proposes LEAVES, a novel module that automatically learns data augmentation methods within a contrastive learning framework applied to biobehavioral time-series data. Conventional contrastive learning relies on data augmentation techniques, but finding optimal augmentation methods and parameters is difficult and time-consuming. LEAVES learns augmentation hyperparameters within a contrastive learning framework using adversarial learning. Experimental results on various biobehavioral datasets using SimCLR and BYOL demonstrate competitive performance compared to existing methods, significantly improving efficiency with a significantly smaller number of parameters (approximately 20) than existing methods (e.g., ViewMaker). LEAVES requires virtually no manual hyperparameter tuning, making it suitable for large-scale or real-time medical applications.

Takeaways, Limitations

•
Takeaways:
◦
We provide an efficient and effective contrastive learning framework for bio-behavioral time series data.
◦
Achieve high performance with a small number of learnable parameters, reducing computational costs.
◦
Increases practicality by reducing the need for manual hyperparameter tuning.
◦
It has high applicability in large-scale or real-time medical applications.
•
Limitations:
◦
Further research is needed to determine how well the generality of the proposed method extends to other types of time series data or other contrastive learning frameworks.
◦
Only experimental results on various bio-behavioral datasets are presented, and performance evaluations on other types of data are lacking.
◦
There is a lack of analysis of the complexity and interpretability of the LEAVES module itself.
👍
Made with Slashpage